Hashing perceptual para la integridad de imágenes de documentos: un procedimiento experimental

Abstract

In this work, a reproducible experimental procedure is proposed for the empirical evaluation of perceptual hashing algorithms applied to the verification of the integrity of images containing textual documents. The adopted approach is based on simulation through the automatic generation of synthetic images and controlled variants incorporating both semantic and non-semantic modifications, which helps overcome the limitations associated with the use of pre-existing real-world datasets.

Within this experimental setup, several perceptual hashing functions are analyzed by evaluating their behavior using metrics based on the Hamming distance. The results obtained allow the characterization of the performance of these algorithms, revealing difficulties in detecting minimal semantic alterations as well as a considerable sensitivity to global luminance transformations.

Based on these observations, the proposed experimental procedure is intended to serve as a methodological basis that facilitates the development of comparable and extensible studies in similar scenarios.